Probabilistic Map-based Pedestrian Motion Prediction Taking Traffic Participants into Consideration

As pedestrians are one of the most vulnerable traffic participants, their motion prediction is of utmost importance for intelligent transportation systems. Predicting motions of pedestrians is especially hard since they move in less structured environments and have less inertia compared to road vehicles. To account for this uncertainty, we present an approach for probabilistic prediction of pedestrian motion using Markov chains. In contrast to previous work, we not only consider motion models, constraints from a semantic map, and various goals, but also explicitly adapt the prediction based on crash probabilities with other traffic participants. Also, our approach works in any situation; this is typically challenging for pure machine learning techniques that learn behaviors for a particular road section and which might consequently struggle with a different road section. The usefulness of combining the aforementioned aspects in a single approach is demonstrated by an evaluation using recordings of real pedestrians.